基于超声影像纹理分析鉴别诊断乳腺分叶状肿瘤和纤维腺瘤

The ultrasound image texture analysis for phyllodes tumor and fibroadenoma of breast

目的乳腺分叶状肿瘤(PTB)和纤维腺瘤(FB)超声特征具有一定重合。文中探讨超声影像纹理分析鉴别诊断PTB和FB的价值。方法选取53例PTB和114例FB患者的术前超声影像资料。将超声二维影像导入MaZda 4.6软件中,手动勾画病变的感兴趣区(ROI),分别选择Fisher系数、分类错误概率联合平均相关系数(POE+ACC)、交互信息(MI)以及联合3种方法(Fisher+POE+ACC+MI)。选择最具鉴别价值的纹理特征参数,构建人工神经网络模型,比较PTB和FB纹理特征的差异,并评估3种纹理分析方法和超声医师对PTB和FB的误判率。结果3种纹理参数分析方法共选取的30组纹理参数中,8组差异有统计学意义(P<0.05)。在总误判率方面,3种方法联合分析的误判率最低,与Fisher系数、POE+ACC、MI以及超声医师的误判率相比,差异均具有统计学意义(χ2=30.683、7.467、12.371、4.138,P<0.05)。同时,超声医师对PTB的误诊率明显高于FB(54.72%vs 17.54%,P<0.05)。结论超声影像纹理分析可用于鉴别诊断PTB和FB。

Objective To investigate the value of ultrasound image texture analysis for the diagnosis of phyllodes tumor and fibroadenoma of breast.Methods A total of 53 patients withphyllodes tumor of the breast(PTB)andfibroadenoma of the breast(FB)were enrolled in this study.The ultrasound images were imported into Mazda 4.6 software and regions of the interest(ROI)were manually drawn.To build the model of artificial neural network,the optimum texture parameters were selected respectivelyfrom Fisher,probability of classification error and average correction coefficient(POE+ACC),mutual information(MI)and the combination of three methods(Fisher+POE+ACC+MI).The differences ofthe ultrasound image texture analysis for PTB and FB were compared,and the misdiagnosis rates of the four texture parameters and ultrasound doctors were assessed.Results Among the figures of 30 groups selected by three texture parameters,there were statistical differencesin 8 groups(P<0.05).The misdiagnosis rate of the combination of Fisher+POE+ACC+MI was the lowest,and there were significant differences compared with Fisher coefficient,POE+ACC,MI and ultrasound doctors(χ2 values were 30.683,7.467,12.371,4.138,P<0.05).Meanwhile,the misdiagnosis rate of PTB of ultrasound doctors was significantly higher than that of FB(P<0.05).Conclusion Ultrasound imagetexture analysis can be used to diagnose phyllodes tumor and fibroadenoma of breast.